Research on Pilots ’ Mental Workload Classification in Simulated Flight

Xue, Jinna; Wang, Changyuan · 2023 · Crossref

DOI: 10.2478/ijanmc-2023-0048

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the critical safety issue of pilot mental workload in aviation, noting that 60% to 90% of flight accidents are linked to excessive cognitive load or stress. The research aims to develop an effective classification method for mental workload using electroencephalogram (EEG) signals combined with subjective evaluation, thereby improving human-computer interaction safety and reliability in flight operations. The experimental design involved eight male graduate student volunteers aged 22–30, who performed simulated flight tasks using a DCS World platform with a Su-25T fighter aircraft. Participants completed three distinct task conditions: control, low load, and high load. The high-load condition required monitoring eight specific cockpit dial parameters, including airspeed, pitch, altitude, heading, roll, steering compartment status, landing gear, and engine status. The low-load condition required monitoring only three parameters: airspeed, pitch, and altitude. EEG data was collected using an ErgoLAB system with a 10-20 standard lead semi-dry cap. Subjective workload was assessed using the NASA-TLX scale, which rates mental effort across six dimensions on a 0–100 scale. Data processing involved filtering raw EEG signals using Butterworth filters for 50Hz interference removal and 0.05–30Hz bandpass filtering. Feature extraction utilized the discrete Fourier transform to estimate power spectral densities for delta, theta, alpha, and beta frequency bands using the Welch method. The core classification model combined a Convolutional Neural Network (CNN) for local feature extraction with a Long Short-Term Memory (LSTM) network for temporal sequence analysis. The model architecture included two 1D convolutional layers, max pooling, and LSTM layers, trained with an L1-loss function. The dataset was split into 80% for training and 20% for testing, with the model trained for 500 epochs at a learning rate of 0.001. The results demonstrated that the proposed CNN-LSTM model achieved a classification accuracy of 94.9%, outperforming traditional methods. Training accuracy exceeded 90% by the 120th epoch, with loss values converging rapidly. Subjective evaluation via NASA-TLX revealed a statistically significant positive correlation (p<0.01) between task difficulty and workload scores, confirming that higher task complexity increased perceived mental load. The study concludes that the combined CNN-LSTM approach is highly effective for classifying pilot mental workload from EEG signals. This method offers significant practical value for monitoring pilot cognitive states, predicting overload conditions, and optimizing human-computer interaction designs to enhance flight safety.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• workload measurement
• mental demand
• neuro workload indices

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data, self report data
• Theoretical Contribution: theory or model